【作者】 李小泉；

【导师】 高岚；

【作者基本信息】 武汉理工大学 ， 轮机工程， 2006， 硕士

【摘要】 随着近代科学技术,特别是信息科学、材料科学、微电子技术和计算机技术的迅速发展,检测技术所涵盖的内容更加深刻、更加广泛。现代人类的社会生产、生活、经济交往和科学研究都与检测技术息息相关。各个科学领域,特别是生物、海洋、航天、气象、地质、通信、控制、机械、交通和电子等,都离不开检测技术,检测技术在这些领域中也起着越来越重要的作用。因此,检测技术已成为人类社会进步的一个重要基础技术,是各学科高级工程技术人员必须掌握的重要的基础技术。 本文主要利用图像处理、模式识别等技术,进行了缺陷自动检测试验系统的关键技术开发、缺陷图像实时采集处理、图象预处理、特征提取、缺陷的分类等方面的研究。其目的在于促进缺陷检测系统高效、智能化技术的发展。本论文主要工作如下: 1、概述了国内外关于自动检测技术现状,分析了数字图像采集和处理技术现状和趋势,采用了基于TI DSP的数字图像采集和处理系统。 2、根据缺陷图像的特点,采用了图像线性增强、直方图均衡化方法来增强对比度。研究了均值滤波、SUSAN滤波、高斯滤波和中值滤波技术,对缺陷图像进行了平滑去噪处理,最后选用快速中值滤波并取得了良好的效果,为下一步图像分割打好了基础。在对图像预处理的基础上,从经典的边缘提取方法入手,研究了Marr-Hildreth和Canny两种改进的边缘检测方法,其中Canny方法效果最好。研究了几种图像阈值分割算法,OSTU、过渡区、最大熵和熵关联的算法,其中过渡区和熵关联的分割算法取得了良好的分割效果。 3、研究用BP神经网络进行缺陷的分类。在分割的基础上提取了不变矩作为缺陷的特征并对矩特征归一化;最后把提取的矩特征输入BP神经网络进行训练和分类。 本文采用VC对算法进行了缺陷分类的仿真试验,结果表明本文采用的方法是有效可行的。论文还对算法的实时性进行了讨论,表明本文预处理方法实时是可行的,但病害分类方法的实时性有待进一步研究。更多还原

【Abstract】 In recent years, with rapid development of infomation technique 、 material technique、 micro-electronics technique and computer technique, inspection technique is more and more profound and abroad. The production 、 existance、 economy communication and science research of modern human society are all related to inspection technique. Each research domain, especial biolog、 ocea、 communications、 geology 、 spaceflight、 weather、 control、 machine、 traffic are all not independent to inspection technique. So the inspection technique is a very important basic technique in human advanced society.In this paper, we use the image processing, pattern recognition technology to exploit distress automatic inspection system, and carry on distress processing research in order to promote the intelligent technology development of the distress inspection.1. Firstly, the paper summarizes current situation in automatic detection technique both at home and abroad. Secondly, the paper analyzes current situation and trend of digital image capture and processing technology, adopts the digital image process system based on TI DSP.2. According to the characteristic of distress image and uses linear image enhancement, histogram equalization technology to enhance image contrast. The paper studies mean filter, SUSAN filter, median filter and gauss filter technology, and selects the fast median filter to get good smoothing result of pavement image. On the basis of pavement image preprocessing, the paper starts with the classical edge detection method, researches two kinds of improved edge detection methods: Marr-Hildreth and Canny. Then the paper studies image segmentation algorithms include: OSTU, transition region, maxim entropy and entropy correlation algorithm, and transition region, entropic correlation make the good segmentation result.3. The paper uses BP neural network to classify distress. The invariant moments are used to represent pavement distress image feature, and normalization features are inputted BP neural network classifier to train and classify. The experiment resultshows that this classifier can recognize pavement distress better when a little noise exists in pavement image. The paper also discusses the real-time ability of the algorithm, experiment shows that preprocessing algorithm fulfils real-time ability but the real-time ability of the classify method need study further.更多还原